add TensorBoardLogger to transformers integration

[bfineran]

adds a sparseml TensorBoardLogger to the transformers integration to enable tensorboard logging of sparsification modifier values.

implementation is currently setup so that TensorBoardLogger is only created if the default transformers tensorboard logging callback is setup and will share a logging directory with it so the sparsification and default transformers training values will both appear on the board.

default transformers tensorboard logged values include run arguments, epoch, loss, and learning rate.
the tensorboard logs are written to --logging_dir, so the tensorboard server can be launched with

tensorboard --logdir <LOGGING_DIR>

additionally investigating if there is any further information to be gained by setting sync_tensorboard=True in the wandb logger